|In response to "AMD enhances Opteron pitch. But who's buying it?" Troy Frank writes:|
I will just admit up front that I am excited about Opteron, since that will likely become obvious anyway. That said, there are several things I thought that made your article very glossed over and biased towards Intel. I'll get to the main reasons people will buy Intel later.
1. You spend a lot of time intentionally or inadvertently trying to lead people to the Opteron/Itanium comparison, and seem to carefully avoid talking about Xeon, which is the most direct competitor to Opteron and the one that Intel is probably the most worried about (i.e., the cash cow).
Change the comparison to the Xeon, and suddenly the picture changes a lot. The Opteron is faster 70 to 80 percent of the time compared to the Xeon, it's cheaper and you also get the ability to run 64-bit software with the same hardware. When Intel finally caves into 64-bit Xeon extensions (as they've practically admitted they will), it will only make things better for the Opteron. It'll still be faster (integrated memory controller and HyperTransport), cheaper, and Intel's support for a 64-bit Xeon will drive faster 64-bit software adoption. That will be a little embarrassing for Intel, since not that long ago the company was screaming to everyone that 64-bit was unnecessary for all but the biggest iron--something many developers said they were crazy for claiming, but what do those crazy coders know.
2. You say it's faster "in certain situations." That's downplaying it almost to the point of bias right there. Other than the Linpack benchmark, Photoshop, and a couple other static content programs that love huge caches, the Opteron has been shown to have consistently faster performance than the Xeon. That's not counting the average 10 to 25 percent improvement the Opteron will get with the jump to 64-bit software.
Not all software will care, but much of it will. If you doubt this, check Google. Many sites have done extensive server benchmarking that show the Opteron as much faster in file serving, Web serving, databases, and other applications. The Xeon is better for some things, but it's not even close to a 50/50 split.
3. You grant them that the Opteron is cheaper, but don't seem to mention (or care) about that point very much. Instead, you go on and on about how you'd probably just buy a new server anyway in 18 to 24 months when 64-bit software is available. Many IT organizations strapped for cash would probably beg to differ with you, however.
When you can buy a server that's faster in the first place, provides a 64-bit performance boost whenever you need it, and is cheaper, that's going to be very compelling for a lot of people. It also lets them maintain acceptable performance longer, meaning they don't have to buy new servers as often, saving even more money. Remember, the Xeon is mostly slower right now, more expensive and has no option for the 64-bit performance boost. Why would you want to go that way?
4. Regarding the Itanium code porting issue. You say that Giga "found there were no significant costs involved with converting existing 32-bit high level code to a non-x86 64-bit architecture." That's a teeny bit misleading, and a lot missing the point. It's misleading because it doesn't say WHO the costs are insignificant for. For example, suppose it costs $50,000 to convert an application to IA-64 and only $10,000 to convert to x86-64. That additional cost may be insignificant to a Fortune 500 or 1000 company, but it wouldn't be insignificant to the bulk of companies. You may quibble with my exact numbers, but nearly anyone I've heard write or talk about it has said that it is much more expensive to port to IA-64 than to x86-64.
Now to the part that's missing the point. Microsoft's success with backwards compatibility in compilers and OSs, has shown conclusively that developers lean towards the familiar. There are far fewer good compilers for IA-64 than for x86-64. There's an even larger gulf in the numbers of developers working on each platform. Given the choice, which platform are they more likely to migrate to? An extension to the environment that they're familiar with or to a completely new environment that I've heard many describe as more complicated?
So, why are people going to buy Intel anyway?
A. The IBM effect: "No one ever got fired for buying Intel."
B. Marketing: Intel does a lot more of it to cover up the Opteron's advantages.
C. Money: Intel has a lot more of it, and pays off manufacturers when possible with "marketing co-op" money, (e.g., the Intel Inside campaign).
D. Volume: AMD only has one fab right now (with another coming). This is not enough volume to supply a large percentage of the server market.
At best, if AMD sold everything it made, the company might be able to get 15 to 25 percent market share.
In conclusion, there are reasons to stick with Intel (see above). However, those would not be good reasons for a lot of people. Either way, let's at least discuss real reasons instead of continuing this FUD [Fear, Uncertainty and Doubt] campaign that dances around,confusing the issue.